Kalman contrastive unsupervised representation learning
Mohammad Mahdi Jahani Yekta · 2024
Read Kalman contrastive unsupervised representation learning by Mohammad Mahdi Jahani Yekta (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-11618453.txt.
Description
Mohammad Mahdi Jahani Yekta presents a Kalman filter based method for learning image features without labels, in work from twenty twenty-four. The method, called KalCo, looks up entries in a dynamic dictionary of encoded keys. A Kalman filter encoder keeps that dictionary large and steady.
A famous prior method uses a fixed momentum rule and counts as a simple case of KalCo. On a standard image task, KalCo reached far higher accuracy than that prior method on a million image set. Gains held on a billion image social set and on brain scan sets. An upgraded version with a stronger head and more augmentation pushed scores higher still. The author reports wins over three top recent rivals.
Catalog record
| CREATOR | Mohammad Mahdi Jahani Yekta |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 21 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |